Multi-threshold color image segmentation based on region growing

An efficient color image segmentation approach based on region growing is presented, which combines region growing and merging to implement the process. To extract the region seed, the image is first transformed from RGB space to HSV space, the hue value difference between adjacent pixels is calculated, a multi-threshold approach is applied to obtain the seed candidate region, finally region centre pixel is regarded as the region seed point. At region growing phase, a region homogeneity criterion is applied to implement the region growing, the color intensity value in R,G,B channel is regarded as a three-dimension vector, and three vector space angles are calculated corresponding to R,G,B channel respectively. Only the differences between adjacent space vector angles on each channel are smaller than a preset threshold, then the homogeneity condition is met and the adjacent pixel is added. To obtain the region edge, the intensity gradients on R,G,B channel are calculated respectively, and three thresholds are set to obtain the final region edge, which is regarded as the stop conditions while region growing is implemented. The validity of the proposed approach is testified by experiments.